REDUCING BIAS IN PREDICTING POVERTY PERCENTAGE AT THE VILLAGE LEVEL IN BENGKULU CITY USING SMALL AREA ESTIMATION

Abstract

Analysis of poverty rates at the small area level, especially at the village level, often faces obstacles due to limited sample sizes. As a result, the use of direct estimates has large variances, making them less reliable. One solution to this problem is to use the Small Area Estimation (SAE) method. This study aims to apply the Beta-Binomial Small Area Estimation (SAE-BB-APHL) model by utilizing a second-order Laplace approach to estimate fixed effects that were fixed in the previous model, namely SAE-BB-HL. The SAE-BB-APHL model was applied to the case of poverty rates in each village in Bengkulu City. The results show a decrease in Mean Squared Error (MSE) in estimating the poverty rate when using SAE-BB-APHL compared to direct estimation and SAE-BB-HL. The MSE values produced by the direct estimation method in several villages were 0.20638 in village 1, 0.04865 in village 2, and 0.04865 in village 46. Meanwhile, the SAE-BB-APHL method successfully reduced the MSE to 0.00524 in village 1, 0.00493 in village 2, and 0.00630 village 46. In addition, the MSE in SAE-BB-APHL is also smaller than SAE-BB-HL as evidenced in Table 3. This decrease in MSE indicates that bias is reduced by using the SAE-BB-APHL method. Therefore, this approach can be a superior alternative in poverty analysis at the village level. More accurate estimation results are expected to assist the government in designing more targeted poverty alleviation policies.